jrcg0000/fpga-runtime-for-opencl

Intel® FPGA Runtime for OpenCL™ Software Technology

Intel® FPGA Runtime for OpenCL™ Software Technology

Prerequisites

Linux

• Red Hat Enterprise Linux (RHEL)* or Rocky Linux* 8 or 9
• SUSE Linux Enterprise Server (SLES)* or openSUSE Leap* 15
• Ubuntu* 20.04, or 22.04 LTS
• GCC 7.5.0 and higher
• CMake 3.10 and higher
• Ninja 1.8.2 and higher
• Git
• libelf
• zlib (only needed for simulation flow)

Windows

• Windows* 10 (64 bit)
• Windows Server* 2019 or 2022
• Microsoft Visual C++ (MSVC)* 2017 and higher
• CMake 3.10 and higher
• Ninja 1.8.2 and higher
• Git
• Cygwin
• libelf (install with scripts/install_libelf.ps1)
• zlib (only needed for simulation flow, install with scripts/install_zlib.ps1)

Building the Runtime

Build Instructions for Linux and Windows

Perform these steps to build the runtime:

1. Clone and change to the runtime repository.

   git clone https://github.com/intel/fpga-runtime-for-opencl
   cd fpga-runtime-for-opencl
   

2. Switch to the latest release version.

   git checkout v2024.2
   

3. Create and change to the runtime build directory.

   mkdir build
   cd build
   

4. Generate the Ninja build files using the cmake command.

   cmake -G Ninja .. -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=../install
   

5. Build the runtime.

   ninja
   

6. Install the runtime to the runtime installation directory.

   ninja install
   

Build Options

• Build without debugging symbols and with optimization.

  cmake -G Ninja .. -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=../install
  

• Build with debugging symbols and without optimization.

  cmake -G Ninja .. -DCMAKE_BUILD_TYPE=Debug -DCMAKE_INSTALL_PREFIX=../install
  

• Use GCC as the compiler instead of the system default (for Linux only).

  CC=gcc CXX=g++ cmake -G Ninja .. -DCMAKE_BUILD_TYPE=Release -DCMAKE_INSTALL_PREFIX=../install
  

• Specify installation paths to libelf and zlib (for Windows only).

  cmake -G Ninja .. -DCMAKE_BUILD_TYPE=Release -DCMAKE_PREFIX_PATH=C:\libelf;C:\zlib -DCMAKE_INSTALL_PREFIX=../install
  

  Alternatively, specify header and library directories explicitly.

  cmake -G Ninja .. -DCMAKE_BUILD_TYPE=Release -DCMAKE_INCLUDE_PATH=C:\libelf\include;C:\zlib\include -DCMAKE_LIBRARY_PATH=C:\libelf\lib;C:\zlib\lib -DCMAKE_INSTALL_PREFIX=../install
  

Running Unit Tests

To run unit tests, execute the following command in the runtime build directory:

ctest -V

Notes

• On Windows, you need to set the paths to the libelf and (optionally) zlib libraries.

  set PATH=C:\libelf\lib;C:\zlib\lib;%PATH%
  

• Running unit tests is not supported for ARM cross compilation.

• To run unit tests individually, first set the board package directory:

  export AOCL_BOARD_PACKAGE_ROOT="$(git rev-parse --show-toplevel)/test/board/a10_ref"
  

  Switch to the directory containing the test executable:

  cd build/test
  

  Run the tests whose group and name contain the given substrings:

  ./acl_test -v -g acl_mem -n create_buffer
  

Using the Runtime

With IntelⓇ FPGA SDK for OpenCL™ or IntelⓇ FPGA RTE for OpenCL™

Perform these steps if you have the IntelⓇ FPGA SDK for OpenCL™ or IntelⓇ FPGA RTE for OpenCL™ installed on your system:

1. In the IntelⓇ FPGA SDK for OpenCL™ installation directory, source the initialization script.

   • On Linux:

     source init_opencl.sh
     

   • On Windows:

     call init_opencl.bat
     

2. Update the /etc/OpenCL/vendors/Altera.icd file with either the filename libalteracl.so if LD_LIBRARY_PATH contains the full path of the runtime installation directory; otherwise, the full path of libalteracl.so itself.

3. Compile your OpenCL host program with the OpenCL header files included from the runtime installation directory.

   -Ifpga-runtime-for-opencl/install/include
   

4. Link your OpenCL host program with the libOpenCL.so library from the IntelⓇ FPGA SDK for OpenCL™ installation directory.

   -L$INTELFPGAOCLSDKROOT/host/linux64/lib -lOpenCL
   

5. Run your OpenCL host program.

Without IntelⓇ FPGA SDK for OpenCL™ or IntelⓇ FPGA RTE for OpenCL™

Perform these steps if you do not have the IntelⓇ FPGA SDK for OpenCL™ or IntelⓇ FPGA RTE for OpenCL™ installed on your system:

1. Download, build, and install the OpenCL ICD Loader.

2. Create the /etc/OpenCL/vendors/Altera.icd file with either the filename libalteracl.so if LD_LIBRARY_PATH contains the full path of the runtime installation directory; otherwise, the full path of libalteracl.so itself.

3. Compile your OpenCL host program with the OpenCL header files included from the runtime installation directory.

   -Ifpga-runtime-for-opencl/install/include
   

4. Link your OpenCL host program with the libOpenCL.so library from the OpenCL ICD Loader.

   -lOpenCL
   

5. Run your OpenCL host program.

Debug Environment Variables

Environment Variable Title	Description
ACL_DEBUG	Set this variable to a value of 1 to 4 to output general-purpose debug messages from the OpenCL Runtime.
ACL_HAL_DEBUG	Set this variable to a value of 1 to 5 to increase debug output from the hardware abstraction layer (HAL), which interfaces directly with the MMD layer.
ACL_CONTEXT_CALLBACK_DEBUG	Set this variable to a value of 1 to produce error text for errors that occur in the OpenCL context. This is error text that would be sent upstream through a callback function provided with clCreateContext.

Notes

• When setting the environment variable OCL_ICD_FILENAMES for debugging, ensure it doesn't resolve to any libalteracl.so other than the one specified in the /etc/OpenCL/vendors/Altera.icd file. Mismatches may lead to multiple runtime library instances being linked into the program, causing undefined behaviour.